Storm-0940 uses ARTEX AI agents to compromise ConnectWise and Microsoft 365 tenants

A China-state aligned threat actor tracked as Storm-0940 used an agentic AI penetration-testing framework to accelerate a supply chain attack involving ConnectWise Cloud Backup, Microsoft said in an analysis published this week. The attackers used compromised application credentials associated with the service to attempt to access the Microsoft 365 environments of more than 2,250 downstream customer tenants. In multiple instances, the attackers obtained privileged access to an organization’s cloud infrastructure in less than three hours after initial targeting, Microsoft reported [1, 2, 3].

Microsoft researchers observed that the activity demonstrates how threat actors are integrating agentic AI into existing offensive security workflows to scale operations exponentially. While the majority of the targeting was not successful, Microsoft said Storm-0940 achieved privileged cloud persistence at many organizations in a brief time [4, 5, 6].

The ARTEX automation

Storm-0940 used an AI penetration-testing framework called ARTEX to automate and accelerate its operations, according to Microsoft. ARTEX is a Go-based, multi-agent large language model (LLM) system that can use any OpenAI-compatible model backend to coordinate a team of agents. These agents can run autonomous penetration tests from reconnaissance through data exfiltration with little human interaction, Microsoft explained. The framework uses a dual-graph multi-agent architecture where a Planner agent maintains a shared task list and Worker agents execute specific subtasks, such as reconnaissance, credential testing, or privilege escalation [2, 5, 7].

The threat actor began using ARTEX as early as September 2026 to target hundreds of organizations globally across sectors including aerospace, defense, energy, and financial services. Between October 10 and October 14, 2026, Microsoft observed Storm-0940 conducting an automated cross-tenant password-spray campaign. This campaign targeted 327 accounts across eight organizations and generated over 151,000 failed logon events. The activity peaked on October 11 at 2,900 sign-in attempts per minute, a pace Microsoft noted leaves defenders with very little time to detect and respond [7, 8, 9].

On October 15, 2026, the attackers shifted their focus to the ConnectWise supply chain. After gaining control of the identities associated with multiple ConnectWise cloud applications, the AI agents used the stolen credentials to authenticate to downstream Microsoft 365 customer environments. This method allowed the attackers to bypass multi-factor authentication (MFA) entirely, as they exploited OAuth 2.0 application permissions that had already been pre-authorized by the victims [2]. Because the applications were already trusted by the tenants, the agents could operate without requiring an interactive user sign-in [2, 7, 8].

Exploiting the OAuth trust model

The attackers leveraged the broad permissions granted to the ConnectWise Cloud Backup service, which was formerly known as SkyKick. To perform its function of backing up and restoring data in Exchange Online, SharePoint, and Teams, the application requires high-level access to a customer's Microsoft 365 tenant. Microsoft noted that the agents used these pre-granted permissions to execute Microsoft Graph operations, which is the programmatic interface used to access data and settings within Microsoft 365 [1, 2, 7].

Once authenticated, the AI agents performed directory reconnaissance and attempted to modify privileges to ensure they could maintain access. Microsoft observed the agents making application credential changes and attempting to alter identities, groups, and privileged roles. These modifications are particularly dangerous because they can create persistent backdoors that remain active even after the original compromised application is remediated or its credentials are revoked [3, 7, 8].

The scale of the automated attempt was significant, with Microsoft reporting that ARTEX was used to attempt sign-ins across more than 2,250 tenants. Huntress, a security research firm, reported seeing sign-in activity from the malicious infrastructure across more than 1,000 Microsoft tenants, noting that more than 300 of those attempts appeared to succeed [10]. However, Huntress Principal Security Operations Analyst Jonathan Semon cautioned that in some of those 300 cases, the ConnectWise application was not actually present in the environment, meaning a "successful" sign-in attempt does not always equate to successful access to a victim's data [3, 6, 10].

Disputed impact and technical limits

While the scale of the authentication attempts is clear, the exact number of organizations that suffered a full data breach remains unknown. Microsoft has not disclosed how many customers were successfully compromised or whether any data was actually exfiltrated, adding that a successful login does not necessarily indicate that customer data was accessed or stolen [1, 5, 6].

While Microsoft and other researchers have described the tool as enabling "agentic" and "autonomous" operations, some experts suggest the term may be overstating the reality of the intrusion [2, 11]. Alan Woodward, a security expert from Surrey University, noted that while automation can significantly scale scanning, password attempts, and email extraction, the available technical material does not prove that the AI agents carried out the entire intrusion from start to finish without human oversight. He suggested the activity might be better described as AI-assisted automation rather than fully autonomous hacking [11].

Furthermore, the origin of the initial ConnectWise compromise remains unestablished. While Microsoft has linked the subsequent activity to Storm-0940, ConnectWise has not publicly explained how the attackers first gained control of the application identities associated with its Cloud Backup product. This leaves a critical gap in the timeline, as it is unclear whether the attackers exploited a vulnerability in the ConnectWise software itself or utilized other means to steal the service's administrative credentials [3, 4, 9].

The challenge of machine-speed defense

Yih Khai Wong, a senior research manager at IDC Asia/Pacific, stated that the timeline of gaining privileged access in under three hours is beyond the reach of any human analyst queue, regardless of how well-staffed a security operations center (SOC) might be. Wong argued that because the attack moves at a pace humans cannot match, containment actions such as revoking tokens and isolating service principals must be designed to trigger automatically based on predefined thresholds [3].

The incident also highlights a structural vulnerability in how organizations manage third-party trust. Yash Kadakia, founder and CTO of Security Brigade, observed that many enterprises focus heavily on securing employee logins with multi-factor authentication while continuing to implicitly trust third-party applications. He noted that these applications can hold equally powerful permissions and authenticate without any employee being involved, effectively creating a gap in the zero-trust model where machines are trusted more than humans [3].

This shift toward commoditized, AI-driven tooling means that high-level operational sophistication is no longer reserved for well-resourced nation-state actors. Wong observed that because attackers are increasingly turning to open-source frameworks like ARTEX, anyone capable of running a GitHub repository against an LLM API can now launch complex, large-scale operations. This democratization of offensive capability suggests that the volume and velocity of attacks will likely continue to increase [1, 3, 5].

Guidance for recovery and investigation

Defenders are being urged to take specific steps to secure their environments as the recovery process continues. Microsoft recommends that customers review every Microsoft 365 tenant that might have been tied to ConnectWise and advises them to preserve all logs and forensic evidence before making any changes to the environment. Microsoft also suggests using its Defender XDR tools to hunt for suspicious Microsoft Graph activity and known malicious threat infrastructure [4, 5, 7].

ConnectWise has advised its customers to keep affected application connections offline until further notice. The company is working to restore its services in phases and has provided a technical investigation and remediation guide to help customers review their individual Microsoft 365 environments. To assist with this, ConnectWise Chief Technology Officer David Raissipour said the company's security organization is walking managed service providers through the process of using Microsoft Entra ID to look for indications of compromise and unauthorized elevated privileges [10, 11].

Ross Filipek, CISO at Corsica Technologies, added a layer of caution regarding the reconnection of services. He stated that before reconnecting with ConnectWise, affected organizations must have a clear understanding of what may have occurred inside their specific environments. He pointed to the discovery of application credential changes as a primary risk, noting that organizations need to investigate whether unauthorized changes were made that could provide attackers with a way back into the system after the initial connection is restored [4].

ConnectWise is currently conducting a "full-court press" to determine if there was any actual access to data from Microsoft 365 tenants, a process that company officials say is ongoing alongside their restoration efforts [2, 9, 10].

References

  1. [cyberscoop] Microsoft pins ConnectWise supply-chain attack to China-state aligned threat group (2026-10-21 · threat intel report · RL-I-2026-0844) · archived
  2. [www.techtimes.com] China's AI Agent Hacked 2,250 Microsoft 365 Tenants in Hours; MFA Didn't Apply (2026-10-22 · threat intel report · RL-I-2026-0844) · archived
  3. [www.csoonline.com] China-linked hackers used agentic AI in attack on 2,250 Microsoft 365 tenants (2026-10-22 · threat intel report · RL-I-2026-0844) · archived
  4. [darkreading] Microsoft Links Storm-0940 to ConnectWise Supply Chain Attack (2026-10-22 · threat intel report · RL-I-2026-0844) · archived
  5. [microsoft_security_blog] AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise (2026-10-21 · threat intel report · RL-I-2026-0844) · archived
  6. [the_record] Microsoft blames ConnectWise attacks on Chinese state hacker using ARTEX AI agents (2026-10-22 · threat intel report · RL-I-2026-0844) · archived
  7. [thehackernews] China-Linked Hackers Use ARTEX AI Agents in ConnectWise Supply Chain Attack (2026-10-22 · threat intel report · RL-I-2026-0844) · archived
  8. [sentinelone] The Good, the Bad and the Ugly in Cybersecurity - Week 43 (2026) (2026-10-23 · threat intel report · RL-I-2026-0844) · archived
  9. [theregister_security] Microsoft says ConnectWise breach let Chinese spies loose on 2,250 cloud tenants (2026-10-22 · threat intel report · RL-I-2026-0844)
  10. [www.crn.com] China State-Aligned Group Storm-0940 Used Agentic AI At Rate Of 2,900 Sign-In Attempts Per Minute In ConnectWise Breach (2026-10-23 · threat intel report · RL-I-2026-0844) · archived
  11. [thehackacademy.com] Microsoft's Storm-0940 report: automation, not AI autonomy (2026-10-24 · threat intel report · RL-I-2026-0844) · archived

This article was created in its entirety by artificial intelligence. No humans were involved in its production. The story was chosen by software, written by gemma4-26b-a4b and checked sentence by sentence against its sources by qwen3.8-27b-q5-nothink.

Every feature · this article as JSON

This page is drawn in the browser. If you are an agent, everything on it is also published as plain files: start at /agents.md. Keep each source's evidence_class attached to anything you repeat. Reuse is unrestricted (CC BY 4.0): credit The Red Lens with a link.

The Red Lens · Newsletters · Blog · About · For agents